Jiangliu Wang (王江柳)

I am a postdoctoral researcher at CUHK T Stone Robotics Institute, where I work on building machines that understand the real world from videos with minimal supervision.

I obtained my PhD degree from The Chinese University of Hong Kong (CUHK) in 2020, where I was advised by Prof. Yun-hui Liu and funded by HKPFS. I obtained my bachelor degree from Nanjing University in 2015, where I was advised by Prof. Wei Li. During my PhD, I did an internship at Tencent AI Lab, where I had a wonderful time working with Dr. Jianbo Jiao, Dr. Linchao Bao, Dr. Wei Liu, and other labmates.

I am seeking collaborations. If you are interested in learning video representations, please feel free to email me.

CV  /  Google Scholar  /  Github  /  Email

profile photo
Research

I am interested in learning representation from videos (along with audio as free resources) with minimal supervision. Recently, I also develop an interest in fundamental machine learning theory.

Self-supervised Video Representation Learning by Uncovering Spatio-temporal Statistics
Jiangliu Wang* , Jianbo Jiao*, Linchao Bao, Shengfeng He, Wei Liu, Yun-hui Liu
T-PAMI, 2021
pdf / early access / code / bibtex

This work is an extension of our CVPR 2019 paper. I recommend the ablation study section. Our approach achieved decent performance after just one training epoch.

Learning to Identify Correct 2D-2D Line Correspondences on Sphere
Haoang Li, Kai Chen, Ji Zhao, Jiangliu Wang, Pyojin Kim, Zhe Liu, Yun-hui Liu
CVPR, 2021
pdf / bibtex

Line correspondences are mapped into vectors tangent to sphere. Neighboring vectors mapped from inliers exhibit a local trend consistency (analogous to “a school of fish”).

Self-Supervised Video Representation Learning by Pace Prediction
Jiangliu Wang, Jianbo Jiao, Yun-hui Liu
ECCV, 2020
pdf / short video / long video / code / bibtex

This work is inspired by the observation that human visual system is sensitive to video pace, e.g., slow motion, a widely used technique in film making.

Self-supervised spatio-temporal representation learning for videos by predicting motion and appearance statistics
Jiangliu Wang , Jianbo Jiao, Linchao Bao, Shengfeng He, Yun-hui Liu, Wei Liu
CVPR, 2019
pdf / code / bibtex

Neural networks are asked to predict motion and appearance statistics, including the largest motion area, largest and smallest color diversity areas.

View-invariant human action recognition based on a 3d bio-constrained skeleton model
Qiang Nie, Jiangliu Wang, Xin Wang, Yun-hui Liu
TIP, 2019
pdf / bibtex

A 3D bio-constrained skeleton model is proposed to recover the corrupted skeletons and encode the body-level motion features into images.

Kinematics features for 3D action recognition using two-stream CNN
Jiangliu Wang , Yun-hui Liu
WCICA, 2018
pdf / bibtex

Temporal encoded kinematics features are proposed for action recognition, which compute the linear velocity and orientation displacement based on human skeleton data.

Robot Intelligence for Real World Applications
Yun-hui Liu, Fan Zheng, Ruibin Guo, Jiangliu Wang, Qiang Nie, Xin Wang, Zerui Wang,
Chinese Journal of Electronics, 2018
pdf / bibtex

This is an editor invitation paper. A brief review is presented to introduce our recent works on machine intelligence for real-world applications of robots. One technology leads to a startup company VisionNav.

A child caring robot for the dangerous behavior detection based on the object recognition and human action recognitions
Qiang Nie, Xin Wang, Jiangliu Wang, Yun-hui Liu
ROBIO, 2018
pdf / bibtex

A caring robot is developed to detect dangerous behavior of children in the domestic environment based on action recognition and object recognition technologies.

Before PhD

During my undergraduate study, I was lucky enough to work with Prof. Wei Li on a defense–intrusion interaction optimization problem. This work is published in a Tier 1 applied mathematics journal.

Motion patterns and phase-transition of a defender–intruder problem and optimal interception strategy of the defender
Jiangliu Wang, Wei Li
Communications in Nonlinear Science and Numerical Simulation, 2015
pdf / bibtex

An optimal interception strategy of the defender is provided with interpretations of its physical meaning, which depends on relative mobility of the intruder and defender.

Thesis
Self-Supervised Video Representation Learning
Jiangliu Wang
September, 2020
pdf  /  slides

Service

Reviewer of CVPR 2021, ICCV 2021, ICML 2021 Workshop on SSL, NeurIPS 2020 Workshop on SSL, ICRA 2020, IROS 2019, T-NNLS, RAM.


Page template from Jonathan T. Barron.